1 Introduction

1.1 Learning Objectives

By the end of this tutorial, you will be able to:

  • Load and visualize Landsat satellite imagery in R
  • Create RGB composite images from individual spectral bands
  • Crop and subset raster data to areas of interest
  • Apply different contrast enhancement techniques
  • Create false-color composites for different applications
  • Calculate vegetation indices (NDVI)
  • Perform change detection analysis

1.2 Background

Landsat satellites have been continuously monitoring Earth’s surface since 1972, providing invaluable data for environmental monitoring, land use planning, and change detection. Landsat 8, launched in 2013, captures data across 11 spectral bands at 30-meter resolution.

In this tutorial, you’ll work with Landsat 8 imagery of the Aarhus region in Denmark, learning how to process and visualize satellite data using R’s raster package.

Note: If you encounter errors with plotRGB() and the scale parameter, ensure your raster package is up to date (update.packages("raster")). Some older versions may have issues with large scale values.

1.3 Setup and Package Installation

1.3.1 Installing Required Packages

The raster package is the primary tool for working with raster data in R. You can also use its newer iteration, the terra package, but we’ll focus on raster for this tutorial.

# Install the raster package (only needed once)
#install.packages("raster")

1.4 Loading Libraries

# Load the raster package
library(raster)

Question 1: What package provides the sp (spatial) classes that raster depends on? (Hint: check the startup message)

raster loaded the required package ‘sp’.

1.5 Understanding Raster Structures

The raster package provides three different structures depending on your data:

Function Use Case
raster() One file, one band
brick() One file, multiple bands (e.g., RGB images)
stack() Multiple files, multiple bands (e.g., Landsat GeoTIFFs)

For Landsat data, we typically use stack() because each spectral band is stored in a separate file.

2 Creating RGB Raster Visualizations

2.1 Loading Individual Bands

Landsat 8 visible light bands: - Band 2: Blue (0.45-0.51 μm) - Band 3: Green (0.53-0.59 μm) - Band 4: Red (0.64-0.67 μm)

# Load the Blue Band (Band 2)
blue <- raster("Data/LC08_L1TP_196021_20190629_20200827_02_T1_B2.TIF")

# Load the Green Band (Band 3)
green <- raster("Data/LC08_L1TP_196021_20190629_20200827_02_T1_B3.TIF")

# Load the Red Band (Band 4)
red <- raster("Data/LC08_L1TP_196021_20190629_20200827_02_T1_B4.TIF")

print(red)
## class      : RasterLayer 
## dimensions : 378, 421, 159138  (nrow, ncol, ncell)
## resolution : 30, 30  (x, y)
## extent     : 567615, 580245, 6217755, 6229095  (xmin, xmax, ymin, ymax)
## crs        : +proj=utm +zone=32 +datum=WGS84 +units=m +no_defs 
## source     : LC08_L1TP_196021_20190629_20200827_02_T1_B4.TIF 
## names      : layer 
## values     : 6339, 29529  (min, max)

Question 2: Use the print() function on one of your raster objects. What is the resolution (pixel size) of Landsat 8 data?

The resolution is 30 by 30 meters.

2.2 Stacking Bands

# Combine the three bands into a single RasterStack
rgb.Scene <- stack(red, green, blue)

2.3 Plotting RGB Composite

Landsat 8 Level 1 data uses 16-bit values ranging from 0 to 65535. We need to specify this range when plotting.

Note: If you get an error about alpha levels, try using stretch = "lin" instead of the scale parameter.

# Plot the RGB composite
plotRGB(rgb.Scene,
        r = 1, # Index of the Red channel in the RasterStack
        g = 2, # Index of the Green channel in the RasterStack
        b = 3, # Index of the Blue channel in the RasterStack
        scale = 65535, # Maximum value for 16-bit data
        axes = TRUE, 
        main = "Landsat Scene - Full Extent"
)

Question 3: Why does the image appear very dark?

The scale is so large, and the reflectance is quite low from the Earths surface, so the distribution in the layers are in the low/dark range.

2.4 Cropping to Area of Interest

Landsat scenes cover over 30,000 km², which can make it difficult to see local features. Let’s crop to the Aarhus region.

2.4.1 Defining the Boundary

# Define a raster with the boundary coordinates
# Coordinates are in latitude/longitude (WGS84)
boundary <- raster(
  ymx = 56.20,  # Maximum y coordinate (top border)
  xmn = 10.09,  # Minimum x coordinate (left border)
  ymn = 56.10,  # Minimum y coordinate (bottom border)
  xmx = 10.29   # Maximum x coordinate (right border)
)

2.4.2 Projecting the Boundary

The boundary needs to match the coordinate system of the Landsat data.

# Project the boundary to match the Landsat scene's CRS
boundary <- projectExtent(
  object = boundary,
  crs = rgb.Scene@srs  # Extract CRS from the Landsat stack
)

2.4.3 Cropping the Scene

# Crop the Landsat image to the Aarhus region
AarhusReg <- crop(
  x = rgb.Scene,      # The raster to crop
  y = boundary       # The extent/boundary to crop to
)

# Plot the cropped RGB composite
plotRGB(AarhusReg,
        r = 1, g = 2, b = 3,
        scale = 65535,
        axes = TRUE, 
        main = "Landsat - Aarhus Region"
)

Note that the displayed image looks the same, since we already cropped the image in advance, so that you can load it on PositCloud. The original image was too large to load into memory.

Question 4: Does the cropped image still appear dark? Why?

Yes, it still appears dark. It is beacuse the scale is still very large, and the data distribution of the layers are the same as before.

3 Improving Visualization with Contrast Stretching

When pixel values occupy only a small portion of the possible value range (0-65535), images appear dull: either very dark, bright or gray, lacking detail in color space. Contrast stretching redistributes values across the full display range. In an 8-bit image, this range is 0-255 and the ‘plotRGB()’ function assumes 255 as the maximum value by default. If ‘scale’ is defined, it is used as the new maximum value. If ‘stretch’ is defined, it will rescale the values ignoring ‘scale’ or the default maximum value of 255.

3.1 Linear Stretch

Linear stretching rescales values evenly from minimum to maximum.

# Apply linear stretching
plotRGB(AarhusReg,
        r = 1, g = 2, b = 3,
        scale = 65535,
        stretch = "lin",  # Type of stretch
        axes = TRUE,
        main = "Landsat - Aarhus\nLinear Stretch"
)

3.2 Histogram Stretch

Histogram stretching distributes values more evenly across the range based on their frequency. It applies a histogram equalization, which means that it stretches dense parts of the histogram and condenses sparse parts.

# Apply histogram stretching
plotRGB(AarhusReg,
        r = 1, g = 2, b = 3,
        scale = 65535,
        stretch = "hist",  # Type of stretch
        axes = TRUE,
        main = "Landsat - Aarhus\nHistogram Stretch"
)

Question 5: Compare the two stretching methods. Which produces better results for this scene?

I assess the “linear” stretching to be the best, as the plot with “histogram” stretching appears to bright so that it is hard to see details, probably because there a many pixels with buildings, roads and concrete that reflects more compared to plants.

3.3 Custom Stretch Using Percentiles

For more control, we can manually define the stretch based on data percentiles.

3.3.1 Exploring Data Distribution

Use the histogram function in R to explore data distributions.

# Plot histograms for each band
hist(AarhusReg[[1]])  # Red band

hist(AarhusReg[[2]])  # Green band

hist(AarhusReg[[3]])  # Blue band

3.3.2 Calculating Percentiles

Derive percentiles of the data distributions at 0%, 1%, 5%, 25%, 50%, 75%, 95%, 99% and 100% of the data. For example, the 75th percentile means that 75% of data points lie below this value.

# Calculate percentiles per band, for all bands
qs_ls <- quantile(AarhusReg, c(0, 0.01, 0.05, 0.25, 0.5, 0.75, 0.95, 0.99, 1))

# Extract and store the extreme values (low and high) across bands for key percentiles:
min0 <- min(qs_ls[, "0%"])
min5 <- min(qs_ls[, "5%"])
max95 <- max(qs_ls[, "95%"])
max100 <- max(qs_ls[, "95%"])

3.4 Applying Custom Stretch

# Plot with custom stretch based on the 5th and 95th percentiles of the data distribution
plotRGB(
  AarhusReg - min5,           # Subtract minimum to shift range
  scale = max95 - min5,         # Set scale to 90th percentile range
  zlim = c(0, max95 - min5),  # Limit display range
  axes = TRUE,
  main = "Landsat - Aarhus\nCustom Stretch"
)

Question 6: Why do we subtract min5 or min0 from the raster before plotting?

It is to shift the data to the left of the distribution, so that the plot isn’t too bright.

Extra Question (advanced): What is a key difference between the custom stretch and the built-in image stretch option?

The blue band has the highest reflectance (because of the atmosphere), and that is corrected for by the former methods. The custom methods shows the true relationship between the three bands, so it appears more bluish.

3.5 Gamma Correction (Advanced)

Gamma correction applies non-linear transformations to the data to remove data skewness, and therefore to ensure a more perceptually even representation of colors. The goal is to transform the data so that it is unskewed and closer to a normal distribution.

# Extract individual bands
blue_band <- AarhusReg[[3]]
green_band <- AarhusReg[[2]]
red_band <- AarhusReg[[1]]

# Rescale values to 0-1 range based on the 5th and 95th percentile for each band
max_range <- qs_ls[, "95%"] - qs_ls[, "5%"]
blue_band <- (blue_band-qs_ls[3, "5%"])/(max_range[3])
green_band <- (green_band-qs_ls[2, "5%"])/(max_range[2])
red_band <- (red_band-qs_ls[1, "5%"])/(max_range[1])

# Set negative values to zero
blue_band[blue_band < 0] <- 0
green_band[green_band < 0] <- 0
red_band[red_band < 0] <- 0

# Set values above 1 to 1
blue_band[blue_band > 1] <- 1
green_band[green_band > 1] <- 1
red_band[red_band > 1] <- 1

# Plot histograms before correction
hist(blue_band)  # Blue band

hist(green_band)  # Green band

hist(red_band)  # Red band

# Apply gamma transformation: use exponent 0.5 for a square-root transformation
blue.Gamma <- blue_band^0.5     # Emphasize mid-tones with square-root transformation
green.Gamma <- green_band^0.5   # Emphasize mid-tones with square-root transformation
red.Gamma <- red_band^0.5       # Emphasize mid-tones with square-root transformation

# Plot histograms after correction
hist(blue.Gamma)  # Blue band

hist(green.Gamma)  # Green band

hist(red.Gamma)  # Red band

# Stack the gamma-corrected bands
rgb.Gamma <- stack(red.Gamma, green.Gamma, blue.Gamma)

# Plot
plotRGB(rgb.Gamma,
        scale = 1,
        zlim = c(0, 1),
        axes = TRUE,
        main = "Landsat - Aarhus\nGamma Correction"
)

Note that there are many different ways to transform your data based on the original data distribution and the assumptions for your statistical analysis and data visualization. For example, you can look into log-transformation or the box-cox transformation that transforms data into a shape close to a normal distribution.

3.6 Using Pre-processed Natural Color Images

Landsat provides pre-processed “Natural Color” GeoTIFF files that require less work. They are meant for visualization purposes only and for data exploration, providing an image data stretch that is close to our human perception.

# Load the Natural Color Image (all bands in one file)
rgb.Nat.Color <- brick("Data/LC08_L1TP_196021_20190629_20200827_02_T1_refl.TIF")

# Crop to Aarhus region
rgb.Nat.Color <- crop(
  x = rgb.Nat.Color,
  y = boundary
)

# Plot
plotRGB(rgb.Nat.Color,
        axes = TRUE,
        main = "Landsat - Aarhus\nNatural Color Image"
)

Question 7: How does this compare to your manually created RGB composites?

It appears much green and red. Looks like the blue band values have been shifted to the low end of its distribution, so that the plot either appears red or green, or the mix that is brown.

4 False-Color Composites

Beyond so-called “true-color” RGB images, different band combinations using different wavelengths highlight different surface features. Below you will explore a few common false-color composites. Landsat bands are provided in brackets, e.g. (2) for band 2.

4.1 Color Infrared (NIR Composite)

A very common false-color composite uses the near-infrared band assigned to the red color, the red band as green and the green band as blue (Landsat band numbers in brackets):

Bands: NIR (5), Red (4), Green (3)
Purpose: Vegetation health analysis

Healthy vegetation appears bright red because it strongly reflects near-infrared radiation. You can also find more information about Landsat bands and common false-color composites here.

# Load and crop near-infrared band
nir <- raster("Data/LC08_L1TP_196021_20190629_20200827_02_T1_B5.TIF")
nir <- crop(x = nir, y = boundary)

# Create Color Infrared stack
rgb.ColInf <- stack(
  nir,           # NIR as red
  AarhusReg[[1]], # Red as green
  AarhusReg[[2]]  # Green as blue
)

# Plot
plotRGB(rgb.ColInf,
        r = 1, g = 2, b = 3,
        scale = 65535,
        stretch = "lin",
        axes = TRUE,
        main = "Landsat - Aarhus\nColor Infrared"
)

4.2 Short-Wave Infrared

Bands: SWIR-2 (7), SWIR-1 (6), Red (4)
Purpose: Vegetation and soil analysis

# Load and crop SWIR bands
swir1 <- raster("Data/LC08_L1TP_196021_20190629_20200827_02_T1_B6.TIF")
swir1 <- crop(x = swir1, y = boundary)

swir2 <- raster("Data/LC08_L1TP_196021_20190629_20200827_02_T1_B7.TIF")
swir2 <- crop(x = swir2, y = boundary)

# Create SWIR stack
rgb.SwInf <- stack(swir2, swir1, AarhusReg[[1]])

# Plot
plotRGB(rgb.SwInf,
        r = 1, g = 2, b = 3,
        scale = 65535,
        stretch = "lin",
        axes = TRUE,
        main = "Landsat - Aarhus\nShort-Wave Infrared"
)

4.3 Agriculture Composite

Bands: SWIR-1 (6), NIR (5), Blue (2)
Purpose: Crop monitoring

# Create Agriculture stack
rgb.Agro <- stack(
  swir1,           # SWIR-1 as red
  nir,           # NIR as green
  AarhusReg[[3]]  # Blue as blue
)

# Plot
plotRGB(rgb.Agro,
        r = 1, g = 2, b = 3,
        scale = 65535,
        stretch = "lin",
        axes = TRUE,
        main = "Landsat - Aarhus\nAgriculture"
)

Question 8: In the Agriculture composite, what color indicates healthy vegetation?

NIR is set to green color. High NIR reflectance indicate healthy vegetation, so green colours on the map indicate healthy vegetation.

5 Calculating Vegetation Indices

5.1 Understanding NDVI

The Normalized Difference Vegetation Index (NDVI) is a proxy for vegetation greenness and health:

\[NDVI = \frac{NIR - Red}{NIR + Red}\]

  • Range: -1 to +1
  • Water/Bare soil: < 0.2
  • Sparse vegetation: 0.2 - 0.5
  • Dense vegetation: > 0.5

Important: Use Level 2 Surface Reflectance data (L2SP), not raw Level 1 data, for accurate indices.

5.2 Computing NDVI

# Load surface reflectance bands (Level 2)
red_sr <- raster("Data/LC08_L2SP_196021_20190629_20200827_02_T1_SR_B4.TIF")
nir_sr <- raster("Data/LC08_L2SP_196021_20190629_20200827_02_T1_SR_B5.TIF")

# Crop to area of interest
red_sr <- crop(x = red_sr, y = boundary)
nir_sr <- crop(x = nir_sr, y = boundary)

# Calculate NDVI using raster algebra
NDVI <- (nir_sr - red_sr) / (nir_sr + red_sr)

# Set negative values to NA (water, clouds)
NDVI[NDVI < 0] <- NA

# load viridis color palettes
#install.packages("viridis")
library(viridis)

# Define color palettes
colors_vir <- viridis(256)
colors_divergent <- colorRampPalette(c("brown4", "white", "darkcyan"))(255) # used later for difference image

# Plot NDVI
plot(NDVI,
     zlim = c(0, 0.6),
     col = colors_vir,
     colNA = "black",
     main = "NDVI - Aarhus (2019)"
)

Question 9: What do the different colors represent in the NDVI map?

Bluish colours represent low NDVI, i.e., low NIR and/or high Red -> low greenness. Greenish color represent the opposite.

5.3 Change Detection: NDVI Comparison

Let’s compare NDVI between 1989 (Landsat 5) and 2019 (Landsat 8). What do you expect to see?

I expect lower NDVI in and around Aarhus, because the city has expanded.

5.3.1 Loading Historical Data

# Load 1989 Landsat 5 bands (note different band numbers!)
# Landsat 5: Band 3 = Red, Band 4 = NIR
red_sr.1989 <- raster("Data/LT05_L2SP_196021_19890626_20200916_02_T1_SR_B3.TIF")
nir_sr.1989 <- raster("Data/LT05_L2SP_196021_19890626_20200916_02_T1_SR_B4.TIF")

# Crop to boundary
red_sr.1989 <- crop(red_sr.1989, boundary)
nir_sr.1989 <- crop(nir_sr.1989, boundary)

5.3.2 Calculating 1989 NDVI

# Calculate NDVI for 1989
NDVI.1989 <- (nir_sr.1989 - red_sr.1989) / (nir_sr.1989 + red_sr.1989)

# Remove negative values
NDVI.1989[NDVI.1989 < 0] <- NA

5.3.3 Computing Change

# Calculate NDVI difference
NDVI.difference <- NDVI-NDVI.1989

# Plot 2019 NDVI
plot(NDVI,
     zlim = c(0, 0.6),
     col = colors_vir,
     colNA = "black",
     main = "NDVI 2019"
)

# Plot 1989 NDVI
plot(NDVI.1989,
     zlim = c(0, 0.6),
     col = colors_vir,
     colNA = "black",
     main = "NDVI 1989"
)

# Plot difference
plot(NDVI.difference,
     zlim = c(-0.4, 0.4),
     col = colors_divergent,
     colNA = "black",
     main = "NDVI Difference\n[2019 - 1989]"
)

Question 10: What patterns do you observe in the NDVI difference map? What might explain these changes?

There are more red colours on the map, apart from white that indicates no change. The red colour indicates a higher NDVI in 1989 in that particular 30 by 30 m pixel. This is mostly around the city, and can be explained by building buildings on farmland. A few places are bluish indicating higher NDVI in 2019, which in most cases can be explained by newly planted forest on farmland.